Skip to content

Network Cost Calculation

This page explains, in plain language, how the web app estimates network costs between model regions.

The calculation runs automatically after model regions are finalized in Step 1 (automatic clustering or manual finalization). If successful, results are stored in app state and exported in Step 9 under extra_inputs/ with a descriptive generated filename, such as extra_inputs/network_costs_10r_eastern-western_nercr.csv.

For overall workflow context, see Web Application.

Plain-English Method

The core assumption is simple: each model region represents one or more major population centers, and the network cost between regions should reflect what it takes to connect those centers.

The method has two parts:

  1. Between-region connection

    • For each region pair, identify plausible links between major population centers in the two regions.
    • Use the lowest-cost available cross-region link as the interregional component.

  2. Within-region backbone (intra-regional adder)

    • A region can contain multiple major population centers.
    • The model builds a simplified internal network to represent the cost of connecting those centers to each other.
    • This internal cost is added on both ends of an interregional connection (start region and destination region).

So, the total for a route from region A to region B is:

  • the best available A-to-B cross-region link
  • plus A's internal connection adder
  • plus B's internal connection adder

This produces consistent region-level values for:

  • connection cost
  • line losses
  • distance proxy

The goal is not to replicate a full transmission expansion model. The goal is to provide a practical, transparent proxy that scales with user-defined model regions.

Example

Suppose region SouthEast contains Atlanta and Charlotte, and region MidAtlantic contains Philadelphia and Washington D.C.

To estimate the cost of a new transmission path from SouthEast to MidAtlantic:

  1. The app identifies Atlanta and Charlotte as the major population centers in SouthEast, and Philadelphia and Washington D.C. as those in MidAtlantic.
  2. It checks available cross-region network links that touch any of those four cities and picks the cheapest one — say, the Charlotte-to-Washington D.C. corridor.
  3. It then calculates SouthEast's internal adder: the estimated cost of moving power from Atlanta to Charlotte (so Atlanta can reach Charlotte, then onward to Washington D.C.).
  4. It calculates MidAtlantic's internal adder: the cost of connecting Philadelphia and Washington D.C. internally.
  5. The final interregional estimate is Charlotte–Washington D.C. link cost + SouthEast internal adder + MidAtlantic internal adder.

This approach means that a region's total network cost to reach any other region reflects both the direct cross-region link and the cost of aggregating power from across that region.

Purpose

The network cost calculation produces one canonical row per unordered region pair with:

  • Interregional transmission upgrade proxy cost
  • Interregional line losses
  • Interregional distance proxy
  • Intraregional "adder" terms for both start and destination regions
  • Total metrics used for scenario inputs and downstream analysis

This is a heuristic approximation intended to provide consistent, model-region-specific network cost signals after regional aggregation.

How Major Population Centers Are Chosen

Each model region gets a set of "major" metro areas (MSAs):

  • Default rule: keep MSAs above a population threshold (pop_threshold, default 1,000,000).
  • Fallback rule: if no MSA clears the threshold, use the largest MSA in that region.

This ensures every region can participate in the network calculation, including less-populous regions.

How Intra-Regional Adders Are Built

Within each model region, the app creates a simplified internal network among major MSAs:

  1. Build a graph using valid within-region links between those major MSAs.
  2. Find least-cost paths among major MSA pairs.
  3. Build a minimum spanning tree to avoid over-counting redundant routes.
  4. Compute population-weighted average internal cost/loss/distance.

Those weighted averages become that region's intra-regional adders.

If a region has too little internal structure (for example, only one major MSA or no valid paths), intra-regional adders remain zero.

For each region pair in the topology:

  1. Find candidate links that connect major MSAs across the two regions.
  2. Choose the single lowest-cost candidate.
  3. Keep that link as the interregional component for the canonical pair row.

If no valid candidate exists for a region pair, that row is omitted.

Total Metrics

For each output row, totals are the interregional value plus start-region and destination-region adders:

  • total_interconnect_cost_mw = interconnect_cost_mw + start_intraregion_cost_mw + dest_intraregion_cost_mw
  • total_line_loss_frac = line_loss_frac + start_intraregion_loss_frac + dest_intraregion_loss_frac
  • total_mw-km_per_mw = mw-km_per_mw + start_mw-km_per_mw + dest_mw-km_per_mw

What Gets Exported

When Step 9 Download All is used:

  • settings/*.yml are always included.
  • extra_inputs/emission_policies.csv is included only when ESR policies exist.
  • extra_inputs/network_costs_<regions>r_<interconnections>_<grouping>.csv is included only when network costs were successfully computed and stored in state. The generated filename is inserted into the ZIP under extra_inputs/.

Technical Reference

The sections below summarize data dependencies and implementation details for developers.

Inputs and Dependencies

Required network data files

The calculator uses preprocessed files from:

  • web/data/network_data/nodes.csv
  • web/data/network_data/edges.parquet
  • web/data/network_data/topology_base.csv

At runtime in the web app, these are fetched once and cached.

Required settings fields

The calculation depends on:

  • model_regions
  • region_aggregations

Region mapping is built from these fields. If no aggregation is provided, each listed region maps to itself.

Optional settings fields

  • network_lines: Additional region pairs to force into topology consideration. Each pair is treated as an unordered connection and canonicalized to one output row.

Key parameter

  • pop_threshold (default: 1,000,000): Threshold used to identify "major" MSAs for intraregional and interregional calculations.

Implementation Sequence

The implementation is in web/calc_network.py, function calculate_network_from_frames.

  1. Apply model-region mapping:
  2. Map base regions in nodes, edges, and topology to model regions using model_regions + region_aggregations.
  3. Drop rows that cannot be mapped.
  4. Add optional topology pairs:
  5. If settings.network_lines is present, map each pair to model regions.
  6. Canonicalize the pair so only one row is kept per connection.
  7. Select major MSAs in each region:
  8. Major MSA set = MSAs with population >= pop_threshold.
  9. Fallback: if none meet threshold, use the single largest-population MSA for that region.
  10. Compute intraregional adders per region:
  11. Build a within-region subgraph among edges whose endpoints are major MSAs in that same region.
  12. For each pair of major MSAs, find the minimum-cost path (cost-weighted shortest path).
  13. Create an MSA-level graph from those pairwise links.
  14. Compute the minimum spanning tree (MST) on that MSA-level graph.
  15. Compute population-weighted average cost/loss/distance over MST links to get intraregional adders.
  16. Compute interregional component per topology pair:
  17. For each canonical topology row (start_region, dest_region), evaluate candidate edges connecting major MSAs across the two regions.
  18. Choose the single candidate edge with minimum cost.
  19. Store the result under a deterministic region ordering so reverse duplicates are not emitted.
  20. Assemble totals:
  21. Total metrics are interregional value plus start-region intraregional adder plus destination-region intraregional adder.

Output Schema

Output rows are canonical unordered region pairs. start_region and dest_region are stored in deterministic sorted order so the CSV contains only one row for each connection.

Column Description Units
start_region Source model region label n/a
dest_region Destination model region label n/a
start_id Selected source substation ID for interregional link integer ID
dest_id Selected destination substation ID for interregional link integer ID
interconnect_cost_mw Cost of cheapest interregional candidate edge $/MW
line_loss_frac Loss fraction of cheapest interregional candidate edge fraction
mw-km_per_mw Distance proxy for interregional candidate edge MW-km/MW
start_intraregion_cost_mw Intraregional cost adder for source region $/MW
dest_intraregion_cost_mw Intraregional cost adder for destination region $/MW
start_intraregion_loss_frac Intraregional loss adder for source region fraction
dest_intraregion_loss_frac Intraregional loss adder for destination region fraction
start_mw-km_per_mw Intraregional distance adder for source region MW-km/MW
dest_mw-km_per_mw Intraregional distance adder for destination region MW-km/MW
total_interconnect_cost_mw Interregional + source adder + destination adder cost $/MW
total_line_loss_frac Interregional + source adder + destination adder loss fraction
total_mw-km_per_mw Interregional + source adder + destination adder distance MW-km/MW
total_interconnect_annuity_mw Annualized cost using 4.4% WACC and 60-year lifetime $/MW/yr
dollar_year Dollar-year for cost figures year (2018)

Edge Cases and Fallback Behavior

  • Regions without any MSA above the threshold use their largest MSA as fallback.
  • If a region has fewer than two major MSAs, intraregional adders for that region remain zero.
  • If no valid within-region paths exist among major MSAs, intraregional adders remain zero.
  • If no candidate interregional edge exists for a topology pair, that pair is omitted from output.
  • If mapped topology is empty, topology is inferred from observed cross-region edges and canonicalized to one row per pair.
  • Unmapped rows (nodes/edges/topology) are dropped during region mapping.
  • Topology self-loops are removed.

Note

In the web app, if this calculation fails, a warning is shown and the app continues; no network cost CSV is exported for that session.

Limitations and Caveats

  • This is a heuristic approximation, not a full transmission expansion optimization.
  • Results depend on the quality and assumptions of preprocessed network data (nodes.csv, edges.parquet, topology_base.csv).
  • Interregional links are represented by a single cheapest candidate edge per canonical region pair.
  • Intraregional adders are based on major-MSA filtering plus MST aggregation, which intentionally simplifies full network detail.